AARHUS
UNIVERSITY
Department of Biology
A brief introduction to
climate data
HOW TO GET THE MOST OUT OF YOUR CLIMATE DATA
Richard Davy, Senior researcher
Nansen Centre, Bergen, Norway
AARHUS
UNIVERSITY
Department of Biology
Overview
Climate data
•Types of climate data (pros and cons)
•What data is available
Downscaling Methods
•Dynamic vs. Statistical Downscaling
•Kriging
•Co-variates: best practice guidelines
Climate projection & prediction
•How is it made and where can I get it?
•Uses and mis-uses of climate model data
2
AARHUS
UNIVERSITY
Department of Biology
Climate data –what is out there?
Gridded Observations:
•Single-location measurements
•Aggregated and interpolated to fit grid
Shortcomings:
•Limited coverage
•Subject to biases (systematic, drift)
•Uncertainties due to
Conversion from satellite data
Homogenization process
Interpolation methods
•Inconsistencies between different datasets
3
Name
Spatial
resolution
Temporal
resolution
Period
CRUTv4
1
5
o x 5o
Mont
hly
1850
- present
GHCNv3
1,2
5
o x 5o
Monthly
1880
–2016
NOAA
1
5
o x 5o
Monthly
1880
- present
GISTEMP
1
2
o x 2o
Monthly
1880
- present
GHCN_CAMS
1
0.5
o x 0.5o
Monthly
1948
- present
UDelaware
1,2
0.5
o x 0.5o
Monthly
1900
-2017
WorldClim2
1,2,
*
0.042
o x 0.042o
Monthly
1960
-2018
CPC Global
1,2
0.5
o x 0.5o
Daily
1979
-2010
NOAA_land
2
0.5
o x 0.5o
Monthly
1948
- present
GPCP
2
2.5
o x 2.5o
Monthly
1979
- present
GPCC
2
0.5
o x 0.5o
Monthly
1891
- present
Examples of datasets for temperature1and precipitation2
WorldClim temperature station dataWorldClim temperature station data
AARHUS
UNIVERSITY
Department of Biology
What is a reanalysis?
Optimum combination of dynamical model
and observations.
Reanalysis procedure
1) Take wide variety of observations
2) Run forecast model
3) Constrain model using data assimilation
Differences in:
•Observations assimilated
•Assimilation method
•Skill of forecast model
4
AARHUS
UNIVERSITY
Department of Biology
Climate data –what is out there?
Climate Reanalyses:
•Synthesis of models and observations
•Product of choice for climate scientists
doing process studies
Improvements:
•All climate parameters are available from
a single product
•Dynamically consistent
•No spatial or temporal gaps
•Close match to a multitude of observations
•Massive advancement in temporal resolution
5
Name
Spatial
resolution
Observations
assimilated
Temporal
resolution
Period
ERA5
-
Land
0.
1ox 0.1o
G
round-
based
&
S
atellite-
born
Hourly &
monthly
1950*
–present
ERA5
0.28
ox
0.28
o
JRA55
1.25
o x 1.25
o
6 hourly
&
daily
1958
–2012
NCEP
2.5
o x 2.5o
1979
- presemt
MERRA
-
2
0.5
ox 0.5o
H
ourly &
3 hourly &
6 hourly
1980
- present
ERA
-
20C
0.25
o x 0.25
o
Surface
pressure
3 hourly
&
6 hourly
1900
-2010
Examples of global climate reanalyses
Higher-resolution climate reanalyses are available for some
regions
* currently, ERA5-Land data is available going back to 1981
(~5 km or similar) for
e.g. Europe, Arctic
AARHUS
UNIVERSITY
Department of Biology
Essential Climate Variables (ECVs)
https://gcos.wmo.int/en/essential-climate-variables
Relevant, Feasible & Cost effective
Atmosphere & surface
Air temperature; Wind speed and direction;
water vapour; precipitation; surface radiation
budget
Terrestrial
Lakes; Snow cover; Permafrost; Albedo;
Soil moisture; Soil carbon; River discharge;
Water use; Groundwater; Glaciers and Ice caps;
6
AARHUS
UNIVERSITY
Department of Biology
Downscaling –the resolution you need
Dynamical
www.cordex.org host large
ensembles of dynamical downscalings of
Global models. Caution! Regional models
have their own biases –combining these
with uncertainty from global models…
Statistical
Hybrid statistical-dynamical
7
AARHUS
UNIVERSITY
Department of Biology
Statistical –for known relationships
Dynamical
Statistical
Use known relations between
variables e.g. soil moisture
and slope steepness/soil
properties.
Predict values (with uncertainty)
at finer scales based upon these
co-variates at finer resolutions
Hybrid statistical-dynamical
8
AARHUS
UNIVERSITY
Department of Biology
Hybrid –costly, but effective
Dynamical
Statistical
Hybrid statistical-dynamical
Sparse observations of
temperature, wind
Relationships between
stations change with
e.g. wind direction
Use dynamical model to
provide external drift
9
AARHUS
UNIVERSITY
Department of Biology
Kriging –preserves uncertainty
Statistical method
Exploits local relationships
with co-variates
Retains uncertainty
We made a Global 1km*1km
SAT climatology with uncertainty
–you can too!
10
AARHUS
UNIVERSITY
Department of Biology
Co-variates –known strong relationships
Strong relationship between Temperature and Elevation at almost all
temporal resolutions (hourly -> climatology)
11
AARHUS
UNIVERSITY
Department of Biology
Co-variates –known weak relationships
Weak relationship between Precipitaion and Elevation at all
temporal resolutions (hourly -> climatology)
Needs atmospheric dynamical model!
12
AARHUS
UNIVERSITY
Department of Biology
Alternative - soil moisture
Stronger relationship between Soil moisture and Elevation, soil properties,
Slope steepness
13
AARHUS
UNIVERSITY
Department of Biology
Kriging –Co-variate effect on estimate
14
AARHUS
UNIVERSITY
Department of Biology
Kriging –Co-variate effect on uncertainty
15
AARHUS
UNIVERSITY
Department of Biology
Kriging –Co-variate effect on estimate
16
AARHUS
UNIVERSITY
Department of Biology
Kriging –Co-variate effect on uncertainty
17
AARHUS
UNIVERSITY
Department of Biology
Getting the most from your co-variates
1) Identify known relationships
Temperature: Elevation, Slope steepness, Slope direction
2) Make use of functionality where known
Soil moisture = alpha2+ 1/beta + gamma
3) Include interaction between co-variates (when expected)
Temperature = Elevation * Slope steepness + Slope direction * Slope steepness
4) Check there is substantial variation of the covariate within your domain
5) Check relationship holds on timescale you’re interested in
6) Consider using another variable you already downscaled e.g. Temperature
18
Don't do it Daily Weekly Monthly Annual
Use with caution Surface air temperature
All clear Soil moisture 0.5m and below
Soil moisture 0-0.5m
Radiation variables
Wind speed
AARHUS
UNIVERSITY
Department of Biology
Localisation of kriging –trade-offs
Choice of nmax defines radius from which kriging relations are derived
As nmax increases,
estimate converges uncertainty decreases computation time increases
exponentially
19
AARHUS
UNIVERSITY
Department of Biology
Climate models
20
AARHUS
UNIVERSITY
Department of Biology
Climate models
21
More processes
/components added.
Question driven e.g.
ice sheets -> sea level
rise
Improved resolution
We are here --->
AARHUS
UNIVERSITY
Department of Biology
Climate projections
22
Several well-studied
periods/scenarios with large
multi-model ensembles
Detection /attribution studies
•Is this effect caused by X
climate condition?
•Run your model with and
without X
AARHUS
UNIVERSITY
Department of Biology
Climate projections: uncertainty sources
23
Projection uncertainty = 1st (Scenario spread)
+ 2nd (Model processes)
+ 3rd (Internal variability)
AARHUS
UNIVERSITY
Department of Biology
Climate projections: uncertainty sources
24
Projection uncertainty = 1st (Scenario spread)
+ 2nd (Model processes)
+ 3rd (Internal variability)
AARHUS
UNIVERSITY
Department of Biology
Climate projections: Uses and mis-uses
Synthetic time series drawn from the climatological PDFs
Robust projections of changes in mean climate and variability
Confidence in projections depends upon (1) variable (2) spatial scale (3) temporal scale
25
Mean temperature at continental scale
Precipitation extremes at country scale
AARHUS
UNIVERSITY
Department of Biology
Climate predictions
Projections are a boundary condition problem
how much CO2 is emitted?
Weather forecasts are an initial value problem
If I have good observations today, I can predict
what happens up to 14 days ahead
Our goal
Reduce uncertainty on short lead
times (<30 years)
Initialise ocean, other sources of
predictability
26
AARHUS
UNIVERSITY
Department of Biology
Using climate predictions
Fish forecasts
Species following salinity
gradients
Probablistic forecast up to
5 years out
Seasonal forecasting
Seasonal drought forecast
Heatwave risks
27
AARHUS
UNIVERSITY
Department of Biology
Climate projections
Limited in variables
Quickly assess scale of change, robustness
Bias-correct for your region
•Spatial pattern
•Temporal variability
29
AARHUS
UNIVERSITY
Department of Biology
For all the ECVs, projection, prediction +
https://esgf-node.llnl.gov/search/cmip6/
Different experiments
e.g. 21st century scenarios, decadal predictions,
Paleo-climate
Data sorted by temporal resolution; model; domain
(atmosphere, terrestrial,..)
Point and click interface to select variables,
experiment, resolution…
Download direct (point-and-click) or wget scripts
30
